[kineto] Deprecate ClientTraceActivity and merge it with GenericTraceActivity (pytorch#56743)

Summary:
Pull Request resolved: pytorch#56743

Pull Request resolved: pytorch/kineto#184

as part of the migration to ClientTraceActivity -> GenericTraceActivity, now that all CTA mirrors GTA's data structure, we can safely swap out the symbol name.

Test Plan:
- `buck build kineto`
- sandcastle to catch any other breakage in depdendees

Took before and after of `fastrnns` bench
`buck run mode/opt //caffe2/benchmarks/fastrnns:bench -- --cnns resnet50 --group cnns --nloops 1000`

Before
https://fburl.com/perfdoctor/9n0izgji

{F611729029}

After
https://fburl.com/perfdoctor/h9d9tlmp
{F611725475}

Sample ParamComms traces
https://www.internalfb.com/intern/perfdoctor/trace_view?filepath=tree%2Ftraces%2Fdynocli%2F0%2F1619503816%2F127.0.0.1%2Flibkineto_activities_4003656.json.gz&bucket=gpu_traces

https://www.internalfb.com/intern/perfdoctor/trace_view?filepath=tree%2Ftraces%2Fdynocli%2F0%2F1619503816%2F127.0.0.1%2Flibkineto_activities_4003657.json.gz&bucket=gpu_traces

https://www.internalfb.com/intern/perfdoctor/trace_view?filepath=tree%2Ftraces%2Fdynocli%2F0%2F1619503816%2F127.0.0.1%2Flibkineto_activities_4003658.json.gz&bucket=gpu_traces

Reviewed By: gdankel

Differential Revision: D27353973

fbshipit-source-id: ad9eee27e840d41c4b2d78e12b10510e11232e97

Author: chaekit

torch/csrc/autograd/profiler_kineto.cpp

@@ -58,10 +58,16 @@ struct TORCH_API KinetoThreadLocalState : public ProfilerThreadLocalState {
     if (!ctx) {
       return;
     }
+#ifdef USE_KINETO_UPDATED
+    libkineto::GenericTraceActivity op;
+    op.activityType = libkineto::ActivityType::CPU_OP;
+    op.activityName = std::string(fn.name().str());
+#else
     libkineto::ClientTraceActivity op;
+    op.opType = std::string(fn.name().str());
+#endif
     op.startTime = ctx->startUs;
     op.endTime = getTimeUs();
-    op.opType = std::string(fn.name().str());
     op.device = 0;
     op.correlation = ctx->correlationId;
     // optimization - postpone shapesToStr till finalizeCPUTrace
@@ -126,7 +132,7 @@ struct TORCH_API KinetoThreadLocalState : public ProfilerThreadLocalState {
   void addTraceEvents(libkineto::ActivityTraceInterface& trace) {
     const auto& events = *(trace.activities());
     for (const auto& ev_ptr : events) {
-      // ClientTraceActivity events are already processed
+      // CPU_OP events are already processed
       if (ev_ptr->type() != libkineto::ActivityType::CPU_OP) {
         kineto_events_.emplace_back();
         kineto_events_.back()